linux-kernel - Re: [PATCH v2] perf stat: Append to default list if use -e +event

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2d2c8e32-53f2-560d-3f69-a0ed7c3810e6@linux.intel.com>
Date:   Thu, 21 Jan 2021 14:08:52 +0800
From:   "Jin, Yao" <yao.jin@...ux.intel.com>
To:     Arnaldo Carvalho de Melo <acme@...nel.org>,
        Jiri Olsa <jolsa@...hat.com>
Cc:     jolsa@...nel.org, peterz@...radead.org, mingo@...hat.com,
        alexander.shishkin@...ux.intel.com, Linux-kernel@...r.kernel.org,
        ak@...ux.intel.com, kan.liang@...el.com, yao.jin@...el.com
Subject: Re: [PATCH v2] perf stat: Append to default list if use -e +event

Hi Arnaldo,

On 1/21/2021 8:03 AM, Arnaldo Carvalho de Melo wrote:
> Em Wed, Jan 20, 2021 at 10:25:53PM +0100, Jiri Olsa escreveu:
>> On Mon, Jan 18, 2021 at 12:54:37PM +0800, Jin, Yao wrote:
>>> root@...-ppc:# ./perf stat -e +power/energy-pkg/ -a -- sleep 1
> 
>>>   Performance counter stats for 'system wide':
> 
>>>                2.02 Joules +power/energy-pkg/
> 
>>>         1.000859434 seconds time elapsed
> 
>>> The '+' prefix is printed. So I finally decide not to remove the '+' prefix
>>> in order to keep original behavior.
>   
>> hm, originaly there's no purpose for the '+', right?
>> it seems it's more like bug then anything else
>   
>> you added function to the '+' to add default events to specified event,
>> which I think is good idea, but I don't think we should display the
>> extra '+' in output
> 
> The value would be to stress that that is an event added to the ones
> without the + prefix, i.e. the default ones.
> 
> But by having the command line copied over and the added events at the
> first lines we should have that abundantly clear.
> 
> Also we won't print removed events (using -), is that available already?
> 

Sorry, the '-' support is not available in this patch. Can I do the patch for '+' first and then do 
a follow up patch for '-' at next step?

> Nope:
> 
> [root@...co ~]# perf stat -e -cycles sleep
> event syntax error: '-cycles'
>                       \___ parser error
> Run 'perf list' for a list of valid events
> 
>   Usage: perf stat [<options>] [<command>]
> 
>      -e, --event <event>   event selector. use 'perf list' to list available events
> [root@...co ~]#
> 
> Like with:
> 
> [root@...co ~]# perf stat -d sleep 1
>  >   Performance counter stats for 'sleep 1':
> 
>                0.80 msec task-clock                #    0.001 CPUs utilized
>                   1      context-switches          #    0.001 M/sec
>                   0      cpu-migrations            #    0.000 K/sec
>                  59      page-faults               #    0.073 M/sec
>           2,215,522      cycles                    #    2.757 GHz                      (7.40%)
>           1,073,189      instructions              #    0.48  insn per cycle           (80.59%)
>             203,615      branches                  #  253.392 M/sec
>               8,309      branch-misses             #    4.08% of all branches
>             245,866      L1-dcache-loads           #  305.972 M/sec
>              14,024      L1-dcache-load-misses     #    5.70% of all L1-dcache accesses  (92.60%)
>       <not counted>      LLC-loads                                                     (0.00%)
>       <not counted>      LLC-load-misses                                               (0.00%)
> 
>         1.001887261 seconds time elapsed
> 
>         0.000000000 seconds user
>         0.001413000 seconds sys
> 
> 
> Some events weren't counted. Try disabling the NMI watchdog:
> 	echo 0 > /proc/sys/kernel/nmi_watchdog
> 	perf stat ...
> 	echo 1 > /proc/sys/kernel/nmi_watchdog
> [root@...co ~]# perf stat -e -LLC* -d sleep 1
> event syntax error: '-LLC*'
>                       \___ parser error
> Run 'perf list' for a list of valid events
> 
>   Usage: perf stat [<options>] [<command>]
> 
>      -e, --event <event>   event selector. use 'perf list' to list available events
> [root@...co ~]#
> 
> - Arnaldo
> 

So if we just want to append the default list, we only need to set detailed_run=1, then ideally 
perf-stat will print the default list.

But for now, there are no task-clock, context-switches, cpu-migrations, page-faults, instructions, 
branches and branch-misses displayed.

root@...-ppc:~# ./perf stat -e cycles -d -a -- sleep 1

  Performance counter stats for 'system wide':

        124,178,207      cycles                                                        (80.02%)
          6,444,490      L1-dcache-loads                                               (80.01%)
          1,043,169      L1-dcache-load-misses     #   16.19% of all L1-dcache accesses  (80.02%)
            564,474      LLC-loads                                                     (80.02%)
             49,262      LLC-load-misses           #    8.73% of all LL-cache accesses  (79.92%)

        1.001614947 seconds time elapsed

Do we still need the '+' prefix to add the specified event on top of default list? It looks current 
syntax should already support that feature, but just need to fix some issues.

Thanks
Jin Yao